On-implant data compression

Reducing the neural data stream inside the implant, before it is transmitted off the body, so that the wireless link and its power budget become feasible. Approaches span a spectrum: lossless entropy coding of the digitised signal; lossy but information-preserving feature extraction such as sending only threshold-crossing events, binned spike counts, or spike-band power; and learned or model-based compression tuned to the downstream decoder.

The core tension is between how much the link can carry and how much of the waveform must be preserved for later analysis. Threshold crossings can shrink data by one to two orders of magnitude at modest cost to online decoding, but they discard the spike shapes needed for offline sorting and for tracking drift. On-implant compression must also fit a punishing silicon area, power, and heat budget, which limits how sophisticated the algorithm can be.